1. Field of the Invention
The present invention relates to a body position detecting apparatus.
2. Description of the Related Art
In a conventional body position detecting method, a position of a human body is detected without contacting by using a video image as disclosed, for example, in JP-A-2003-109015. In the method, a body model that schematically represents a human body is superimposed on a body in the video image. Then, a position of a feature point of the body model is detected as a coordinate so as to detect the body part. In the method, the body model is required to be superimposed on each frame of the video image. Thus, the method has an issue that a calculation amount is large.
In view of such circumstances, an exemplary method for reducing the calculation amount is disclosed in Masanobu Yamamoto, “Ubiquitous Motion Capture and Applications Thereof”, an academic conference of “Study of Visual Information for Detecting A Human Body”, at a small hall of a library in Osaka Electro-Communication University, 2004. In the method, an initial position of a body is detected by applying a body model to the body in the video image. After that, a displacement of each body part is detected and the displacement is accumulated to the initial position. Thereby, the calculation amount can be reduced.
In the present case, by accumulating the displacement to the initial position sequentially, a detection error of the displacement may also be accumulated. Thus, the detected position of the body part may be out of an actual position with time. Therefore, the initial position is required for being updated as needed. However, the initial position is difficult to be detected by fitting the body model without a constrained condition.